Skip to main content

Installation

Premora installs with the customer-facing premora CLI, which consumes versioned release artifacts. There are three delivery models — connected pull, on-prem registry pull, and physical USB air-gap — and they share the same bundle structure and install flow, so you can move between them without changing how the product installs.

Before you start, confirm the requirements and run premora doctor.

1. Validate the environment​

premora doctor   # validate local toolchain + cluster connectivity

2. Prepare a values file​

Copy the example values file from the bundle and fill in environment-specific settings — secrets, ingress, and backing-service connection details:

cp values.example.yaml customer-values.yaml
# edit customer-values.yaml

3. Install Core​

Choose the delivery model that matches your egress policy.

Connected (pull from the release endpoint)​

premora core install \
  --catalog https://downloads.premora.ai/catalog.json \
  --values customer-values.yaml

On-prem registry​

Mirror the approved catalog and bundles into your own internal registry first, then point the installer at it. This keeps the supply chain inside your control while remaining a pull model.

Air-gapped (offline media)​

premora core install \
  --offline-media /mnt/usb/premora-offline \
  --values customer-values.yaml

In air-gapped mode, the installer loads image archives from the offline media before applying the Helm release.

Equivalent direct bundle flow​

The CLI wraps a bundle that you can also drive directly:

tar -xzf premora-install-1.0.0.tar.gz
cd premora-install-1.0.0
# verify checksums/signatures, then:
cp values.example.yaml customer-values.yaml
./install.sh --namespace premora --values customer-values.yaml

install.sh runs preflight, optionally loads image archives, locates the packaged premora-platform chart, and applies the release with helm upgrade --install.

Useful dry-run options:

./install.sh --dry-run --values customer-values.yaml      # server-side Helm dry run (needs a cluster)
./install.sh --render-only --values customer-values.yaml  # render locally with helm template (no cluster)

4. Verify the install​

helm list -n premora
kubectl get pods -n premora
kubectl get events -n premora --sort-by=.lastTimestamp

Confirm: Helm release status, pod readiness, ingress/gateway reachability, backing-service connectivity, and core-service health endpoints.

Installing Inference (optional)​

Premora Inference ships as a separate bundle with the same artifact model. The GPU host must be bootstrapped first.

premora inference install --offline-media /mnt/usb/premora-offline
premora inference register export   # produce the registration manifest for Core

Upgrades & support bundles​

premora core upgrade                 # upgrade an existing Core release
premora support collect              # gather logs, status, and versions for support

See Upgrades & support for the full procedure, then continue to First run.